IEEE CBMI 2026 ยท Conference Presentation

VideoHEDGE: Entropy-Based Hallucination Detection for Video-VLMs via Semantic Clustering and Spatiotemporal Perturbations

Sushant Gautam, Cise Midoglu, Vajira Thambawita, Michael A. Riegler, Pรฅl Halvorsen

Simula Metropolitan Center for Digital Engineering, Oslo, Norway


๐Ÿ—“๏ธ Conference
IEEE CBMI 2026
October 2026
๐Ÿ“ Location
Rempart Building
Toulouse, France
๐ŸŽค Presentation
Friday, October 23, 2026
11:45โ€“12:00

Session: Regular session 5 (11:30โ€“13:00) ยท Amphitheater

Conference program: cbmi2026.sciencesconf.org/program/details

๐Ÿงฌ CBMI 2026 Website

๐Ÿ“„ Presentation Slides

๐Ÿ”ฌ Paper & Resources

๐Ÿง  VideoHEDGE

VideoHEDGE: Entropy-Based Hallucination Detection for Video-VLMs via Semantic Clustering and Spatiotemporal Perturbations

A modular framework that extends entropy-based reliability estimation from images to video question answering. Reliability is scored with Semantic Entropy (SE), RadFlag and Vision-Amplified Semantic Entropy (VASE), tested on the SoccerChat benchmark with three 7B Video-VLMs.


Resources
arXiv: 2601.08557 ยท Benchmarking library hedge-bench on PyPI